[OpenAI Launches Control Room, a Centralized Management System Capable of Halting Model Output in 35 Seconds] OpenAI announced on its X platform that it has built a central system called Control Room to track model workloads in its research infrastructure and quickly halt them in case of security vulnerabilities. This system covers all research inference workloads, including agent sampling and data processing for model training and evaluation, and works in conjunction with its misalignment monitoring system. It can stop model output within 35 seconds of making an intervention decision, thus terminating or pausing related tasks. OpenAI states that Control Room is part of its overall solution for securely training and deploying cutting-edge models.